🧠 OpenThaiGPT R1 32b

OpenThaiGPT R1 32b is an advanced 32-billion-parameter Thai language model focused on analytical thinking and reasoning, outperforming larger models such as DeepSeek R1 70b and Typhoon R1 70b despite being less than half their size. The model excels at tasks requiring complex analytical thinking, including mathematics, logic, and coding in the Thai language.

Download the model

OpenThaiGPT R1 32b — Hugging Facehuggingface.co

openthaigpt/openthaigpt-r1-32b-instruct

Highlights

• State-of-the-art Thai language model that outperforms larger models on mathematics and logical reasoning benchmarks
• Explicit reasoning capabilities able to show its thought process step by step
• Significantly smaller size (32b) yet higher performance than 70b models
• Specialized in analytical thinking in Thai, including complex mathematical and logical problems
• High coding performance in both Thai and English

Benchmark results

SkyThought	OpenThaiGPT R1 32b	DeepSeek R1 70b	Typhoon R1 70b
AIME24-TH	56.67	33.33	53.33
AIME24	63.36	53.33	53.33
MATH500-TH	83.8	75.4	81
MATH500	89.4	88.88	90.2
LiveCodeBench-TH	62.16	53.15	47.75
LiveCodeBench	69.67	64.97	54.79
OpenThaiEval	76.05	74.17	77.59
AVERAGE	71.58	63.31	65.42

Technical report

OpenThaiGPT 1.6 and R1 Technical Report — arXivarxiv.org

If OpenThaiGPT has been beneficial for your work, kindly consider citing it as follows:

@misc{yuenyong2025openthaigpt16r1thaicentric,
      title={OpenThaiGPT 1.6 and R1: Thai-Centric Open Source and Reasoning Large Language Models}, 
      author={Sumeth Yuenyong and Thodsaporn Chay-intr and Kobkrit Viriyayudhakorn},
      year={2025},
      eprint={2504.01789},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2504.01789}, 
}

How to use

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "openthaigpt/openthaigpt-r1-32b-instruct"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "กรุงเทพมหานครคืออะไร"
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=8192,
    temperature=0.6
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

vLLM

1. Install vLLM (vllm-project/vllm)
2. Run server

vllm serve openthaigpt/openthaigpt-r1-32b-instruct --tensor-parallel-size 2

• Note, change --tensor-parallel-size 2 to the amount of available GPU cards.

3. Run inference (CURL example)

curl -X POST 'http://127.0.0.1:8000/v1/chat/completions' \
-H 'Content-Type: application/json' \
-d '{
  "model": "openthaigpt/openthaigpt-r1-32b-instruct",
  "messages": [
    {
      "role": "user",
      "content": "กรุงเทพมหานครคืออะไร"
    }
  ],
  "max_tokens": 4096,
  "temperature": 0.6,
  "top_p": 0.95,
  "top_k": 40
}'

GPU memory requirements

Number of Parameters	FP 16 bits	8 bits (Quantized)	4 bits (Quantized)
32b	64 GB	32 GB	16 GB

License

• This model is available for research and commercial use under the terms of the Qwen2.5 license agreement. Please refer to the LICENSE file for more information.

Support

• Official website: openthai.aieat.or.th
• Facebook page: OpenThaiGPT Group
• Discord server for discussion and support: discord.gg/7KDdKkBGUs
• Email: kobkrit@iapp.co.th

Disclaimer: Provided responses are not guaranteed.